Endocrine Abstracts

Objectives: In CogThy, a longitudinal study on Graves’ disease, we previously showed that medial temporal lobe volumes (hippocampus and amygdala) are smaller in hyperthyroid patients than in matched thyroid-healthy participants. Longitudinally, we saw a volume increase in the patients in the sense of a recovery of the brain under treatment. In the present work, we conducted a corresponding analysis on the whole brain.

Methods: Automatic segmentations of T1-weighted images (baseline and 15 months after diagnosis) were carried out with MAPER software and the Hammers Atlas Database, augmented to 120 subcortical and cortical regions. We considered regional context class components (background, cerebrospinal fluid, grey matter (GM), white matter (WM), GM and WM summation (GMWM), total volume). Using a split-sample approach, we separated the exploration (hypothesis generation) from the validation. During the exploration, we identified clusters of regions using prior knowledge (“natural clusters”) and a data-driven approach (affine propagation (AP)). For entities showing notable differences at baseline, we formulated hypotheses and tested them on the validation set, claiming significance for differences with P < 0.05 after accounting for multiple comparisons (Benjamini-Hochberg).

Results: At baseline at diagnosis in severe hyperthyroidism (FT4>50 pmol/l), several entities were significantly different between patients and controls: five individual regions, seven natural clusters, and one AP cluster (cf. Table). None of these entities differed between patients and controls at follow-up, suggesting a recovery effect.Table 1 Adj: Benjamini-Hochberg adjusted. indiv: Individual region. NC: natural cluster. AP: data-driven cluster. Diff %: patient/control volume difference (positive if entity is larger in patients).

Conclusion: Graves’ disease produces a morphometric imprint in the patient’s brain beyond the medial temporal lobe, namely in the frontal lobe. This suggests that imaging biomarkers derived via quantitative image analysis should be further explored fo